Photovoltaic panel inclination angle detection and calculation method based on unmanned aerial vehicle airborne laser radar

Abstract

The invention relates to a photovoltaic panel inclination angle detection and calculation method based on an airborne laser radar of an unmanned aerial vehicle. The method is suitable for the technical field of unmanned aerial vehicle airborne radars. According to the technical scheme, the method comprises the steps that photovoltaic panel three-dimensional point cloud data collected by an airborne laser radar of an unmanned aerial vehicle and the scanning height and the scanning angle of the airborne laser radar at the same moment are obtained; preprocessing the three-dimensional point cloud data of the photovoltaic panel; segmenting the preprocessed three-dimensional point cloud data of the photovoltaic panel, and extracting geometric features of the photovoltaic panel in the three-dimensional point cloud data; and determining the inclination angle of the photovoltaic panel based on the scanning height and the scanning angle of the airborne laser radar, the geometrical characteristics of the photovoltaic panel in the three-dimensional point cloud data and the known actual geometrical characteristics of the photovoltaic panel. Based on the three-dimensional point cloud data of the photovoltaic panel acquired by the airborne laser radar of the unmanned aerial vehicle, the inclination angle of the installation panel in the photovoltaic factory can be quickly and accurately calculated through a data processing means and a machine learning technology.

Description

Technical Field

[0001] The present invention relates to a photovoltaic panel inclination detection calculation method based on a drone-mounted laser radar, which is applicable to the drone-mounted radar technology field. Background Art

[0002] Photovoltaic power generation is an emerging renewable energy generation technology. As an important component of clean energy, it has received widespread attention and promotion. In photovoltaic power generation systems, the inclination of photovoltaic panels is one of the important factors affecting power generation efficiency. Traditional photovoltaic panel inclination measurement methods mainly rely on manual measurement or the use of professional equipment, such as theodolites and total stations, which are time-consuming, labor-intensive and inefficient. Summary of the invention

[0003] The technical problem to be solved by the present invention is: in view of the above-mentioned problems, a photovoltaic panel inclination detection and calculation method based on unmanned aerial vehicle airborne laser radar is provided.

[0004] The technical solution adopted by the present invention is: a photovoltaic panel inclination detection and calculation method based on a drone-mounted laser radar, characterized in that it includes:

[0005] Obtain the 3D point cloud data of the photovoltaic panels collected by the drone-mounted laser radar, as well as the scanning height and scanning angle of the drone-mounted laser radar at the same time;

[0006] Preprocess the 3D point cloud data of photovoltaic panels;

[0007] Segmenting the preprocessed three-dimensional point cloud data of the photovoltaic panel to extract geometric features of the photovoltaic panel in the three-dimensional point cloud data;

[0008] The inclination angle of the photovoltaic panel is determined based on the scanning height and scanning angle of the airborne laser radar, the geometric features of the photovoltaic panel in the three-dimensional point cloud data, and the known actual geometric features of the photovoltaic panel.

[0009] The preprocessing of the photovoltaic panel three-dimensional point cloud data includes:

[0010] The octree sampling method in point cloud downsampling is used to simplify the point cloud data, the SOR filtering algorithm is used to denoise the point cloud data, and the NDT algorithm is used for point cloud registration.

[0011] The step of segmenting the preprocessed three-dimensional point cloud data of the photovoltaic panel and extracting geometric features of the photovoltaic panel in the three-dimensional point cloud data includes:

[0012] A segmentation algorithm combining supervoxel and region growing is used. The supervoxel segmentation algorithm identifies adjacent points in the point cloud and merges them into supervoxels with similar features, thereby generating a set of meaningful three-dimensional regions for segmenting and analyzing the point cloud data. After further feature extraction, the geometric features of the photovoltaic panel are detected.

[0013] The method of determining the inclination angle of the photovoltaic panel based on the scanning height and scanning angle of the airborne laser radar, the geometric features of the photovoltaic panel in the three-dimensional point cloud data, and the known actual geometric features of the photovoltaic panel includes:

[0014] Use the RANSAC algorithm to perform plane fitting on the point cloud data, and determine the plane equation of the photovoltaic panel based on the known actual geometric features of the photovoltaic panel;

[0015] Extract the normal vector from the plane equation of the photovoltaic panel, calculate the angle between the normal vector and the horizontal plane, and thus obtain the inclination angle of the photovoltaic panel;

[0016] The scanning height and scanning angle of the airborne laser radar are used to correct the inclination of the photovoltaic panel.

[0017] Also includes:

[0018] The possible error sources in the inclination angle calculation are analyzed and identified. According to the results of the error source analysis and identification, the correction algorithm for the corresponding error correction is determined to correct the inclination angle calculation result.

[0019] The error sources include scanning angle error, laser foot point positioning error, and airborne platform vibration error.

[0020] The analysis and identification of the scanning angle error includes:

[0021] Set scanning angle It is the angle between the laser beam and the Z axis of the laser radar coordinate system. The deviation between the zero angle of the instantaneous scanning angle and the Z axis is called the index difference, which is represented by ε. Due to the existence of the index difference ε, the actual scanning angle is formed.

[0022] Scanning angle error yes and The difference between the two, let n be the number of laser foot points on a scanning line, the ideal scanning angle of the i-th point Actual scanning angle and scanning angle error It can be expressed by the following formula:

[0023]

[0024] Where: is the actual scanning angle; is the ideal scanning angle; is the scanning angle error; ε is the index difference; n is the number of laser foot points on a scanning line; i refers to the laser foot point serial number.

[0025] The analysis and identification of the laser foot point positioning error includes:

[0026] In a certain time interval t, the plane moves forward, resulting in a motion error coefficient m r , is given by:

[0027]

[0028] Where: Δx, Δy, Δz are the motion error components along the three coordinate axes.

[0029] The analysis and identification of the vibration error of the airborne platform includes:

[0030] Vibration can be classified into periodic vibration, non-periodic vibration and random vibration according to the motion law of system vibration.

[0031] The beneficial effects of the present invention are as follows: the present invention can quickly and accurately calculate the inclination angle of the panels installed in the photovoltaic plant based on the three-dimensional point cloud data of the photovoltaic panels obtained by the drone-mounted laser radar through data processing means and machine learning technology, and has the advantages of high automation, global perspective, high-precision measurement, time and cost saving, and improved safety.

[0032] In the present invention, airborne laser radar scanning can quickly obtain large-area photovoltaic panel point cloud data, and the inclination angle can be automatically calculated with the help of data processing means and machine learning technology. This method can provide a global perspective and a comprehensive understanding of the inclination angle distribution, which is more comprehensive than traditional local measurement methods.

[0033] Compared with traditional manual measurement, the airborne laser radar saves time and cost, improves operation safety, and avoids the risk of high-altitude operations. The photovoltaic panel inclination rapid detection calculation method based on airborne laser radar can quickly, accurately and comprehensively provide inclination information, providing an important reference for panel installation, adjustment and maintenance.

[0034] The present invention obtains point cloud data through airborne laser radar, and uses data processing means and machine learning technology to automatically calculate the panel inclination, avoiding the tediousness and errors of manual measurement and improving calculation efficiency and accuracy. The drone can scan in any weather conditions without waiting for good weather conditions, thus ensuring the continuity and reliability of inclination calculation.

[0035] The present invention can be applied to large-scale photovoltaic power generation systems, and the panel tilt angle monitoring and adjustment of the entire system can be realized through airborne laser radar scanning, which greatly improves the operational efficiency and economic benefits of the system. Compared with traditional manual measurement and monitoring methods, the cost of airborne laser radar scanning and data processing technology is low, which greatly reduces the cost of panel tilt angle calculation and monitoring.

[0036] The specific advantages of the photovoltaic panel tilt detection and calculation method using airborne laser radar are as follows: laser radar can obtain high-precision position information of the panel from three-dimensional space, making it more efficient when covering large-area photovoltaic arrays; laser radar is usually based on optical non-contact measurement, its measurement accuracy is high, and it is not affected by the panel material; drones equipped with laser radar can perform monitoring tasks without contacting the photovoltaic array, while for example, if the tilt sensor is fixed on the panel, it cannot adapt to dynamic changes of the panel, such as changes caused by wind load; in addition to obtaining tilt information, laser radar can also provide other important information of the photovoltaic array.

[0037] The present invention can intelligently adjust the panel inclination angle according to the calculation results, thereby improving the photovoltaic power generation efficiency and system stability, and further reducing the operating cost of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Flow chart of the method in the embodiment.

[0039] Figure 2 The structure block diagram of the UAV equipped with the airborne laser radar in the embodiment.

[0040] Figure 3 : is the relationship between the coordinate error and the flight height H of the drone in the embodiment.

[0041] Explanation of the accompanying drawings: 1-wireless remote control; 2-GPS positioning module; 3-flight control unit; 4-power supply system; 5-inertial measurement unit; 6-single-line laser radar; 7-electronic speed regulator; 8-gyroscope; 9-acceleration sensor; 10-industrial computer; 11-UAV rotor. DETAILED DESCRIPTION

[0042] like Figure 1 As shown, this embodiment is a photovoltaic panel tilt detection calculation method based on a drone-mounted laser radar, which specifically includes the following steps:

[0043] S101. Acquire the three-dimensional point cloud data of the photovoltaic panel collected by the drone-mounted laser radar, and the scanning height and scanning angle of the drone-mounted laser radar at the same time.

[0044] In this embodiment, a drone equipped with an airborne laser radar is used to scan and obtain three-dimensional point cloud data of photovoltaic panels, which can achieve comprehensive coverage of large areas. Before scanning, it is necessary to determine the altitude and flight attitude of the drone according to the geometric characteristics (position, shape, size, etc.) of the photovoltaic panels, and select a suitable shooting angle to ensure that the monitoring points can cover the area to be monitored, and to ensure that factors such as light, weather and wind are taken into account during the scanning process to obtain high-quality data.

[0045] In this example, the hardware connection method of the drone is as follows Figure 2 As shown, it mainly includes a single-line laser radar, an industrial computer, a UAV flight control unit, a GPS global positioning system module, a shock absorber, a power connection line, a data transmission line and a power switch.

[0046] The single-line laser radar is connected to the industrial computer through a high-speed data transmission line; the flight control unit is connected to the GPS global positioning system module through a serial port. The flight control unit contains an inertial measurement system, which is composed of a gyroscope and an acceleration sensor; the GPS global positioning system module is directly fixed on the drone to minimize the impact of vibration on position positioning; the shock absorber is connected to the drone flight control unit through a special mounting bracket; the power cable connects the power supply port of the laser radar and the drone power supply system to ensure stable power supply to the system; the power switch is connected to the drone's power supply system, which is convenient for users to start and stop the laser radar system when needed.

[0047] When the drone is flying, the single-line laser radar is powered by the drone power supply system and starts to emit laser beams for target detection. The data collected by the laser radar is transmitted to the industrial computer through the data transmission line, and the industrial computer processes and stores the data. The inertial measurement system in the drone flight control unit monitors the flight attitude and acceleration information of the drone in real time, which facilitates the subsequent correction of the laser point cloud data. The GPS positioning module collects the position coordinate information of the drone and records it in the flight control unit. The vibration damper is installed on the top of the laser radar to reduce the impact of system vibration on the laser radar ranging accuracy. During the field test, after turning on the power switch, the laser radar starts to collect point cloud information and saves it in the industrial computer.

[0048] S102. Preprocess the three-dimensional point cloud data of the photovoltaic panel, use the octree sampling method in the point cloud downsampling to simplify the point cloud data, use the SOR filtering algorithm to denoise the point cloud data, improve data quality and improve algorithm efficiency; use the NDT algorithm to perform point cloud registration on the acquired data.

[0049] In this embodiment, the octree sampling method in point cloud downsampling is used to simplify the point cloud data, so as to reduce the amount of point cloud data while well preserving the original shape characteristics of the point cloud.

[0050] The octree construction process is optimized to set a minimum depth threshold (4 levels) to prevent overly shallow segmentation. Even if there are not enough objects in some areas that should be split, the depth of the tree will not be less than this threshold, thus avoiding the problem of over-simplification. In addition, the improved octree structure based on the OBB (Oriented Bounding Box) outer bounding box can enclose objects more accurately. The specific approach is: use oriented bounding boxes when splitting nodes. These bounding boxes can be adjusted according to the actual position and direction of the object, so as to more accurately enclose the geometric shape of the object. The OBB of each child node can be rotated to adapt to the actual direction of the object, making the segmentation result more accurate.

[0051] In this example, we first set the threshold for point cloud division, obtain the maximum and minimum values ​​of X, Y, and Z of the point cloud data coordinates, and establish the minimum outer bounding cube of the point cloud data based on these values. The minimum outer bounding cube is divided into 8 sub-cubes. If the number of point clouds in the sub-cube is less than the defined threshold, the division is stopped. Otherwise, it is continued to be divided into 8 sub-cubes. If it is found that the amount of point cloud data allocated to the sub-cube is not zero and is the same as the amount of point cloud data contained in the parent cube, the sub-cube stops being subdivided, because according to the theory of space segmentation, the allocation of the subdivided space must be smaller. If the number is the same, it will cause infinite cutting. Repeat the steps of dividing the minimum outer bounding cube until the maximum recursive depth is reached. Finally, the center points of all sub-spaces are used as the positions of the downsampled point clouds to form the downsampled point clouds.

[0052] Finally, the center points of all subspaces are used as the positions of the downsampled point clouds to form downsampled point cloud data. This method uses OBB for spatial partitioning, which ensures the simplification of point cloud data while retaining the original geometric features of the object, and improves the accuracy and flexibility of spatial segmentation and object encapsulation.

[0053] In this embodiment, the SOR filtering algorithm is used to denoise the point cloud data to improve data quality and algorithm efficiency. The denoising principle of the SOR filtering algorithm is to determine outliers based on the distance distribution characteristics from a certain point to all neighboring points in the point cloud data, thereby removing outliers. The basic rule is to calculate the average distance from the point to the nearest neighboring point for each point in the point cloud data. Assuming that the calculated average distance value satisfies the Gaussian distribution, the shape of the Gaussian distribution is determined by the mean μ and the standard deviation σ. Then calculate the mean μ and standard deviation σ of all these distances to determine the distance threshold. The calculation formulas are:

[0054]

[0055] Thresh_d=μ±t*σ

[0056] Where k is the number of neighboring points, n is the number of point clouds, d is the distance from each point to the neighboring point, t is the standard deviation multiple, Thresh_d is the distance threshold, and when the distance d is greater than Thresh_d, the point is judged as an outlier and removed from the main point cloud.

[0057] In this embodiment, the acquired data is used to perform point cloud registration using the NDT algorithm, and a series of optimization adjustments are made to the NDT algorithm, including an improved spatial segmentation method, dynamic resolution adjustment, enhanced iteration strategy, and parallel computing. The steps of NDT can be divided into: establishing a grid map of the first frame of point cloud, and Newton method iteration to maximize the probability of projecting the second frame of point cloud onto the grid map. The specific algorithm flow is as follows:

[0058] When building the grid map of the first frame of point cloud, a multi-resolution grid is used. A low-resolution grid is first built for rough matching, and then a high-resolution grid is used in the fine matching stage. During the grid segmentation process, the resolution of the grid is dynamically adjusted to decide whether to use a high or low resolution grid based on the local characteristics of the point cloud. Based on the points in the grid, the parameters of the normal distribution probability density function (PDF) of the grid are calculated

[0059]

[0060] in, represents the position of all scan points within this grid, and represents the covariance of the points within the Σ grid.

[0061] Furthermore, each point of the second point cloud is transformed according to the transformation matrix. The point of the second scanned point cloud falls on which grid of the reference point cloud, and the corresponding probability distribution function PDF is calculated

[0062]

[0063] Use the adaptive step size adjustment mechanism to optimize the convergence process of the Newton method. During the optimization process, the step size is dynamically adjusted according to the current convergence state to avoid premature convergence or oscillation. The point cloud calculation tasks are processed in parallel to fully utilize the computing power of multi-core processors.

[0064] Find the maximum likelihood function and get the optimal transformation parameters

[0065]

[0066] The current scan is a point cloud Given a set of scan points X and transformation parameters Let space conversion function Indicates the use of posture transformation Let's move And find the PDF of the transformed point, then the likelihood function that can be maximized is the optimal transformation parameter.

[0067] Maximizing the likelihood is equivalent to minimizing the negative log-likelihood function:

[0068]

[0069] An optimization algorithm is then used to adjust the parameters to minimize the negative log-likelihood function. The NDT algorithm uses Newton's method for parameter optimization.

[0070] S103 , segmenting the preprocessed three-dimensional point cloud data of the photovoltaic panel, and extracting geometric features of the photovoltaic panel in the three-dimensional point cloud data.

[0071] This embodiment adopts a segmentation algorithm that combines supervoxel and region growing. The supervoxel segmentation algorithm can identify adjacent points in the point cloud and merge them into supervoxels with similar features, thereby generating a set of meaningful three-dimensional regions for segmenting and analyzing point cloud data.

[0072] In this example, we first use an energy descent method to segment supervoxels. The idea of ​​segmentation is to assume that there are two supervoxel sets S i and S j , if we consider S i Merge to S j , then the expected energy reduction can be expressed as:

[0073] Δ=λ-c j D(r j ,r i )

[0074] In the formula, λ is a regularization parameter, c j Supervoxel S j The number of points in S i and S j The difference distance between them, r i 、r j S i and S j The representative point in the supervoxel S is i Just accept S j In order to eliminate the robustness effect of parameter λ, the initial value of λ is set to the median of the minimum difference distance between each point and its adjacent points using the following formula, and then the value of λ is doubled in each iteration, which can make the iteration converge quickly.

[0075] λ0=median{min p (D(p,N i ))}|p∈P

[0076] Where N i is the set of adjacent points of point p, and P is the total number of point clouds in the segmented scene. Then, based on the point cloud data after supervoxel segmentation, the regional growing method is used for fusion.

[0077] Since the surface of the panel point cloud data detected in this embodiment is relatively smooth, different normal vector thresholds are selected as conditions for regional fusion. According to the idea of ​​K-means clustering, the normal vector angle and orthogonal distance between adjacent regions are calculated respectively by the following formula to measure the similarity of adjacent supervoxel blocks. If the angle meets the set threshold, the merging operation is performed until the merged region and all its adjacent supervoxels do not meet the conditions, and then the merging steps are repeated until all supervoxels have corresponding divided regions.

[0078]

[0079] In the above formula, is the normal vector of the current clustering area, is the normal vector of the adjacent supervoxel, C a is the centroid of the current clustering area, C b is the centroid of the adjacent supervoxel.

[0080] Furthermore, feature extraction is performed to extract geometric features from each segmented area, such as center position, size, shape, etc.

[0081] S104, determining the inclination angle of the photovoltaic panel based on the scanning height and scanning angle of the airborne laser radar, the geometric features of the photovoltaic panel in the three-dimensional point cloud data, and the known actual geometric features of the photovoltaic panel.

[0082] Use the RANSAC algorithm to fit a plane to point cloud data. First, a certain number of points (usually at least 3 points, used to determine a plane) are randomly selected from the point cloud data. These points are used to preliminarily estimate the parameters of the plane model. Based on the selected points, the plane equation is calculated. The general form of the plane equation is:

[0083] Ax+By+Cz+D=0

[0084] Among them, (A, B, C) is the normal vector of the plane, and D is the distance between the plane and the origin.

[0085] From the three points (x1, y1, z1), (x2, y2, z2), (x3, y3, z3) we can calculate the normal vector of the plane:

[0086]

[0087] Using the estimated plane equation, all points are compared to the plane. The distance from each point to the plane is calculated and checked to see if the distance is within a preset threshold. If the distance is within the threshold, the points are considered to conform to the current estimated plane model.

[0088] The distance from a point to the plane can be calculated as:

[0089]

[0090] Count the number of points that conform to the current plane model. If the number exceeds the set minimum point threshold, the current model is considered valid. Repeat the above steps until the maximum number of iterations is reached or the optimal plane model is found and determined as the final model.

[0091] Extracting the normal vector from the plane equation of a photovoltaic panel Calculate the photovoltaic panel tilt angle θ

[0092]

[0093] Thereby obtaining the inclination angle of the photovoltaic panel.

[0094] Associate the drone parameters and use the scanning height and scanning angle of the airborne lidar for correction. In most cases, the inclination of the photovoltaic panel can accurately reflect the inclination of the panel. The scanning height and angle are mainly used for data calibration and precise positioning to ensure that the spatial coordinate system of the point cloud data corresponds correctly to the actual physical environment. At the same time, in complex environments, the scanning parameters help to align the point cloud data with the actual geographic coordinate system, thereby improving the measurement accuracy.

[0095] S105, performing error analysis and correction on the calculation results to improve the calculation accuracy. Analyze and identify the possible error sources in the inclination angle calculation, determine the correction algorithm for the corresponding error correction based on the results of the error source analysis and identification, and correct the inclination angle calculation results.

[0096] In this example, possible error sources in the inclination angle calculation are analyzed and identified, including scanning angle error, laser foot point positioning error, airborne platform vibration error, etc.

[0097] Scanning angle error analysis and identification: Due to the design and installation of the scanning rotation axis inside the laser radar, there is a certain position deviation from the ideal state, so the actual starting angle of the scan is not zero, and the angle data recorded by the system will have a certain error. It is the angle between the laser beam and the Z axis of the laser radar coordinate system. The deviation between the zero angle of the instantaneous scanning angle and the Z axis is called the index difference, which is represented by ε. Due to the existence of the index difference ε, the actual scanning angle is formed. Scanning angle error yes and The difference between the two, let n be the number of laser foot points on a scanning line, the ideal scanning angle of the i-th point Actual scanning angle and scanning angle error It can be expressed by the following formula:

[0098]

[0099] Where: is the actual scanning angle; is the ideal scanning angle; is the scanning angle error; ε is the index difference; n is the number of laser foot points on a scanning line; i refers to the laser foot point serial number.

[0100] Analysis and identification of laser foot point positioning errors: The main influencing factors include the flight altitude of the drone and the vibration of the drone. When the flight altitude changes, the tilt distance from the sensor to the laser point will also change accordingly. Since the attitude angle error and the scanning angle error are multiplied by the tilt distance, the positioning error increases, so the coordinate error will increase with the increase of flight altitude. The relationship between the coordinate error and the flight altitude of the drone is as follows: Figure 3 As shown, the results show that the error m along the scanning direction X Affected by the change of the UAV's flight altitude, the error m along the height direction Z It is less affected by the change of the flying altitude of the drone. In the actual flight test, due to air disturbance, the aircraft will deviate from the designed route with the change of flight speed and the vibration of attitude. We define the difference between the data obtained in the actual situation and the data obtained in the ideal situation as the motion error. In a certain time interval t, the plane moves forward, resulting in the motion error coefficient m r , is given by:

[0101]

[0102] Where: Δx, Δy, Δz are the motion error components along the three coordinate axes.

[0103] Airborne platform vibration error analysis and identification: Vibration is classified into periodic vibration, non-periodic vibration and random vibration according to the motion law of system vibration. If the time expression of the periodic vibration quantity x(t) is a sine or cosine function, then this type of vibration is called simple harmonic vibration. The typical motion equation of simple harmonic vibration can be expressed as:

[0104]

[0105] In the formula, A represents the amplitude, which indicates the maximum displacement during vibration; ω represents the circular frequency or angular frequency, ω = 2πf; Represents the initial phase. Simple harmonic vibration is the simplest periodic vibration. A periodic vibration can be expanded into a superposition of multiple simple harmonic vibrations using harmonic analysis methods. The common characteristics of periodic vibration and non-periodic vibration are: the system characteristics are certain, and the vibration amount is certain at any moment of vibration, so they both belong to deterministic vibration. However, there is no definite functional relationship between the vibration amount and time of random vibration, so it is unpredictable.

[0106] In this embodiment, according to the results of the error analysis, a correction algorithm suitable for error correction is developed. The correction algorithm can correct the calculation results according to the actual error situation to improve the accuracy and obtain a more accurate inclination calculation result. For example, the zero offset correction is used to correct the zero offset error in the measurement system, or the model parameter calibration is used to calibrate and optimize the parameters of the model to better fit the actual situation, thereby improving the prediction accuracy and reliability of the model.

Claims

1. A photovoltaic panel tilt detection calculation method based on drone-mounted laser radar, characterized in that: include: Obtain the 3D point cloud data of the photovoltaic panels collected by the drone-mounted laser radar, as well as the scanning height and scanning angle of the drone-mounted laser radar at the same time; Preprocess the 3D point cloud data of photovoltaic panels; Segmenting the preprocessed three-dimensional point cloud data of the photovoltaic panel to extract geometric features of the photovoltaic panel in the three-dimensional point cloud data; The inclination angle of the photovoltaic panel is determined based on the scanning height and scanning angle of the airborne laser radar, the geometric features of the photovoltaic panel in the three-dimensional point cloud data, and the known actual geometric features of the photovoltaic panel.

2. The photovoltaic panel inclination detection calculation method based on drone-mounted laser radar according to claim 1 is characterized in that: The preprocessing of the photovoltaic panel three-dimensional point cloud data includes: The octree sampling method in point cloud downsampling is used to simplify the point cloud data, the SOR filtering algorithm is used to denoise the point cloud data, and the NDT algorithm is used for point cloud registration.

3. The photovoltaic panel inclination detection calculation method based on drone-mounted laser radar according to claim 1 is characterized in that: The step of segmenting the preprocessed three-dimensional point cloud data of the photovoltaic panel and extracting geometric features of the photovoltaic panel in the three-dimensional point cloud data includes: A segmentation algorithm combining supervoxel and region growing is used. The supervoxel segmentation algorithm identifies adjacent points in the point cloud and merges them into supervoxels with similar features, thereby generating a set of meaningful three-dimensional regions for segmenting and analyzing the point cloud data. After further feature extraction, the geometric features of the photovoltaic panel are detected.

4. The photovoltaic panel inclination detection calculation method based on drone-mounted laser radar according to claim 1 is characterized in that: The method of determining the inclination angle of the photovoltaic panel based on the scanning height and scanning angle of the airborne laser radar, the geometric features of the photovoltaic panel in the three-dimensional point cloud data, and the known actual geometric features of the photovoltaic panel includes: Use the RANSAC algorithm to perform plane fitting on the point cloud data, the actual geometric features of the photovoltaic panel are known, and the plane equation of the photovoltaic panel is determined; Extract the normal vector from the plane equation of the photovoltaic panel, calculate the angle between the normal vector and the horizontal plane, and thus obtain the inclination angle of the photovoltaic panel; The scanning height and scanning angle of the airborne laser radar are used to correct the inclination of the photovoltaic panel.

5. The photovoltaic panel inclination detection calculation method based on drone-mounted laser radar according to claim 1 is characterized in that: Also includes: The possible error sources in the inclination angle calculation are analyzed and identified. According to the results of the error source analysis and identification, the correction algorithm for the corresponding error correction is determined to correct the inclination angle calculation result.

6. The photovoltaic panel inclination detection calculation method based on drone-mounted laser radar according to claim 5 is characterized in that: The error sources include scanning angle error, laser foot point positioning error, and airborne platform vibration error.

7. The photovoltaic panel inclination detection calculation method based on drone-mounted laser radar according to claim 6 is characterized in that: The analysis and identification of the scanning angle error includes: Set scanning angle It is the angle between the laser beam and the Z axis of the laser radar coordinate system. The deviation between the zero angle of the instantaneous scanning angle and the Z axis is called the index difference, which is represented by ε. Due to the existence of the index difference ε, the actual scanning angle is formed. Scanning angle error yes and The difference between the two, let n be the number of laser foot points on a scanning line, the ideal scanning angle of the i-th point Actual scanning angle and scanning angle error It can be expressed by the following formula: Where: is the actual scanning angle; is the ideal scanning angle; is the scanning angle error; ε is the index difference; n is the number of laser foot points on a scanning line; i refers to the laser foot point serial number.

8. The photovoltaic panel inclination detection calculation method based on drone-mounted laser radar according to claim 6 is characterized in that: The analysis and identification of the laser foot point positioning error includes: In a certain time interval t, the plane moves forward, resulting in a motion error coefficient m r , is given by: Where: Δx, Δy, Δz are the motion error components along the three coordinate axes.

9. The photovoltaic panel tilt angle detection calculation method based on drone-mounted laser radar according to claim 6 is characterized in that: The analysis and identification of the vibration error of the airborne platform includes: Vibration can be classified into periodic vibration, non-periodic vibration and random vibration according to the motion law of system vibration.

Images